relax SVN archive

Both relax and cpmg_fit input and output files for the Fyn SH3 R1rho data have been added.

This is for the 'TP02' model and 'NS R1rho 2-site' models. The cpmg_fit results include source code
modifications to show the differences between the various 'corrections'. The dispersion software
comparison file has been updated to include this data and to show the cpmg_fit verses relax
differences.

Complete reworking of the 'NS R1rho 2-site' dispersion model.

The original code of Nikolai Skrynnikov and Martin Tollinger has been modified to match the
behaviour of Dmitry Korzhnev's cpmg_fit software. The equations from Korzhnev et al., JACS 2005
(http://dx.doi.org/10.1021/ja0446855) have been used for the initial magnetisation and the R1rho'
calculation. All equations have been added to the manual to clarify the model.

Made systemtest test pass on Mac OS 10.9

Speed up of the 'TP02' and 'NS R1rho 2-site' dispersion models.

Repetitive calculations have been minimised.

Updated and reformatted the dispersion software comparison document.

Updated the NESSY results for the truncated CPMG data from Flemming Hansen.

This now uses the data from all residues to allow for a proper error analysis so the results are
comparable to all the other software.

Updated the relax results for the truncated CPMG data from Flemming Hansen.

This includes the new results for the 'MMQ CR72' model. The analysis uses more model nesting. And
the Grace plots now include the interpolation graphs (hence the plots are now bzip2 compressed).

The 'LM63' and 'LM63 3-site' models are now classified as nested in the dispersion auto-analysis.

The grid search for the 'LM63 3-site' model is therefore skipped and the starting parameters for
optimisation are set to those of the optimised 'LM63' model.

Fix for a typo in the previous commit (r21864).

Fix for the relax_disp.plot_disp_curves user function.

The interpolated curves now have all invalid points of 1e100 removed from the graph. This allows
for reasonable graph scaling.